Activities of Daily Living Detection through Energy Consumption Data and Machine Learning to Support Independent...

Area of Science:

• Gerontology and Health Informatics
• Artificial Intelligence and Machine Learning
• Smart Home Technology and Energy Monitoring

Background:

• The growing aging population necessitates advanced solutions for supporting independent living.
• Monitoring daily activities is crucial for elder care and timely intervention.
• Smart home technology offers a non-intrusive method for activity recognition.

Purpose of the Study:

• To investigate the feasibility of identifying Instrumental Activities of Daily Living (IADLs) using power consumption data.
• To develop and evaluate machine learning models for classifying IADLs based on energy usage patterns.
• To assess the potential of energy-based activity recognition for Ambient Assisted Living (AAL) applications.

Main Methods:

• Utilized the REFIT dataset comprising smart plug power consumption data.
• Employed unsupervised K-Means clustering for grouping energy consumption patterns.
• Applied supervised Long Short-Term Memory (LSTM) networks for activity classification and prediction.
• Validated models using Silhouette and Davies-Bouldin indices for clustering, and F1-Score for classification.

Main Results:

• K-Means clustering effectively grouped energy patterns (Silhouette score: 0.88, Davies-Bouldin Index: 0.29).
• LSTM models achieved high accuracy in classifying IADLs over time (F1-Score: 0.99).
• Activities like cooking, cleaning, and entertainment were most accurately identified due to distinct energy signatures.

Conclusions:

• Energy-based activity recognition using smart plugs is a feasible approach for monitoring IADLs.
• This non-intrusive method supports independent aging and has potential in AAL environments.
• Future research should focus on detecting activities without direct energy use and integrating contextual data.